We need to extract data from 2 million receipt photos a month. Send images straight to a multimodal model, or run OCR first and send text?

A design trade-off with cost, accuracy and latency all pulling differently. The interviewer wants a measured comparison on real receipts and a hybrid that sends each image down the cheapest path that works.

Updated Oct 2026 · Original practice questions with worked answers and teaching rubrics. Our editorial policy explains the sources, review process and limits.